AI Safety Talent Network (ASTN)
A career command center for AI safety talent. Users maintain living profiles, receive AI-powered opportunity matching with fit explanations, and get personalized career recommendations.
Live at safetytalent.org.
What it does
ASTN connects three groups: individuals pursuing AI safety careers, local organizations (like BAISH) managing their communities, and opportunity posters looking for matched candidates.
- Smart matching — A two-tier LLM matching engine scores opportunities against user profiles, organized into Great / Good / Exploring tiers with explanations of why each opportunity fits
- AI career advisor — A persistent sidebar agent that helps with profile building, opportunity discovery, and career guidance
- Conversational profile building — Users can build their profile through natural conversation or by uploading a CV, instead of filling out forms
- Opportunity aggregation — Automatically scrapes aisafety.com job boards daily, plus supports manually posted org opportunities
- Organization dashboards — Member directory, engagement scoring, event tracking (Lu.ma sync), program management, and coworking space booking
- Personalized career actions — LLM-generated next steps tailored to each user's profile and goals
Tech stack
| Layer | Technology |
|---|
| Frontend | React 19, TanStack Start/Router/Query, Tailwind v4, shadcn/ui |
| Backend | Convex (database, real-time sync, serverless functions) |
| Auth | Clerk (GitHub, Google, Email+Password) |
| AI | Claude (career advisor, extraction), Gemini Flash (matching), Kimi K2.5 (conversation) |
| Email | Resend with React Email templates |
| Analytics | PostHog |
| Deployment | Vercel |
| Runtime | Bun |
Getting started
Prerequisites
- Bun installed
- A Convex account
- A Clerk application configured with GitHub, Google, and Email+Password providers
Setup
- Clone the repo and install dependencies:
git clone <repo-url>
cd ASTN
vp install
- Create a
.env.local file with your client-side keys:
VITE_CONVEX_URL=<your convex deployment url>
VITE_CLERK_PUBLISHABLE_KEY=<your clerk publishable key>
- Set server-side environment variables in the Convex dashboard:
| Variable | Required | Description |
|---|
ANTHROPIC_API_KEY | Yes | Claude API key for matching, career advisor, and extraction |
RESEND_API_KEY | Yes | Resend key for transactional email |
EIGHTY_K_ALGOLIA_APP_ID | No | 80K Hours job scraping |
EIGHTY_K_ALGOLIA_API_KEY | No | 80K Hours job scraping |
AIRTABLE_TOKEN | No | aisafety.com job board + CRM import |
AIRTABLE_BASE_ID | No | Airtable base for imports |
AIRTABLE_TABLE_NAME | No | Airtable table for imports |
- Start development:
bun run dev
This runs both the Vite frontend and Convex backend concurrently.
Other commands
bun run dev:web # Frontend only
bun run dev:convex # Convex backend only
bun run build # Production build
bun run lint # Type check + lint
bun run format # Prettier format
Project structure
src/
routes/ # TanStack file-based routing (all pages)
components/ # React components organized by feature
agent-sidebar/ # AI career advisor
matches/ # Match cards and detail views
profile/ # Profile editor and upload wizard
org/ # Organization pages and admin
ui/ # shadcn/ui primitives
hooks/ # Shared React hooks
lib/ # Client utilities
convex/
schema.ts # Database schema (all tables + indexes)
matching/ # Two-tier AI matching engine
agent/ # Career advisor agent (tools, prompts, threads)
aggregation/ # Opportunity scrapers (80K Hours, aisafety.com)
enrichment/ # Profile extraction from CV and conversation
engagement/ # LLM engagement scoring
emails/ # React Email templates + send logic
careerActions/ # Personalized career step generation
crons.ts # Scheduled jobs (sync, emails, engagement)
Key routes
| Path | Description |
|---|
/ | Dashboard with matches, career actions, events |
/profile/edit | Profile editor |
/matches | Opportunity matches by fit tier |
/opportunities | Browse all opportunities |
/orgs | Organization directory with map |
/org/:slug/admin | Org admin dashboard |
/admin | Platform admin |
License
AGPL-3.0
